Automatic Classification of Software Requirements using Artificial Neural Networks: A Systematic Literature Review

Abstract

Software requirements classification is a human-intensive task performed during the requirements analysis phase in software development. This literature review analyzes the state-of-the-art of the classification of software requirements using Artificial Neural Networks. Fourteen articles were selected to conduct the review. Sixteen different techniques to classify requirements were identified where, besides artificial neural networks, the most popular are Naive Bayes and the Support Vector Machine. Among the reported Artificial Neural Networks, we identify Convolutional Neural Networks and a Shallow Neural Network. We also found seven approaches that classify functional and non-functional requirements, six that classify only non-functional requirements, and one of them that classifies only functional requirements. The most used metrics to express classification results were accuracy, recall, and F-score. Finally, the results of the classifiers are gathered and reported.